Strictly guard against missing stats ETS table
During a restart it was possible for the dashboard to receive a refresh signal before the stats plugin had finished mounting. This introduces a stricter set of conditionals that will safely degrade stats results back to an empty state when the ets table isn’t available.
Increase the timeout while waiting for config on dashboard mount. The restart period for some apps exceeds 5s and mounting would still cause an error.
Update stale sidebar state and queue counts when there aren’t any jobs matching the given state or queue.
When a queue/state combination had no values, the counts on the sidebar would
“stick” and only show the previous non-zero value. This was most noticeable
with the executing state as it worked through jobs quickly.
Apply access controls on the server side as well as the client. It is no longer possible to use blocked actions by manually restoring button HTML in the browser.
Previously, node and queue information on the dashboard was powered by heartbeat
records stored in the oban_beats table. That is no longer the case and the
dashboard is entirely independent from beats and from Pro.
Check out the Oban v2.6 upgrade guide for a walkthrough on using the Gossip plugin for realtime updates.
To secure the dashboard, or comply with an existing CSP within your application,
you can specify nonce keys for images, scripts and styles. For example, to pull
the asset nonce from a single :my_csp_nonce assignment:
oban_dashboard("/oban", csp_nonce_assign_key: :my_csp_nonce)
Remove dependency on oban_pro (again). The dashboard is driven by events and
data provided by Oban alone.
Bump the Oban dependency ~> v2.6